@tradercjz/dsh-dolphindb
DolphinDB 与 DeepSeek Harness 的集成：WebSocket 执行器、面向模型的查询/写入工具，以及捆绑的 DolphinDB 技能，作为一个可安装的配置档案包
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add github:tradercjz/dsh-dolphindb#822caf41716e8b32bf52bdd1e1464c24335322dfDolphindb 以 @tradercjz/dsh-dolphindb 发布，当前版本为 0.1.0。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
DolphinDB integration for DeepSeek Harness as an installable profile bundle. One package ships the WebSocket executor, the model-facing query/write tools, and the 23 bundled DolphinDB skills.
| Component | Surface |
|---|---|
dolphindb_query | Read-only DolphinDB scripts/SQL (result tables, vectors, matrices, scalars; optional chart; optional server override) |
dolphindb_execute | Write/mutation scripts, gated behind the approval seam with a code preview (optional server override) |
dolphindb_switch | Switch the active server environment; persisted in the dolphindb settings section and hot-applied |
| 23 DolphinDB skills | Bundled SKILL.md trees (syntax, SQL, vectorization, functional programming, data import, factors, ops, Orca DStream, plotting, backtest, FICC pricing, Tushare/CSMAR, …) exposed through the skill tool |
dsh plugin --profile <name> add github:tradercjz/dsh-dolphindb
or add @tradercjz/dsh-dolphindb to a profile's dsh.profile.bundles list after @deepseek-ai/dsh-base.
The executor runs a named server registry. The bundle ships one local server
(127.0.0.1:8848 / admin, credential ref DOLPHINDB_PASSWORD) as the
deployment default. Real deployments live in the dolphindb settings section —
$DSH_HOME/settings.yaml, hot-reloaded, and editable from web Settings →
Plugins once @tradercjz/dsh-client-dolphindb is installed:
dolphindb:
active: prod
servers:
prod:
host: 183.134.101.139
port: 8892
username: admin
passwordRef: DOLPHINDB_PROD_PASSWORD
test:
host: 183.134.101.139
port: 30006
username: admin
passwordRef: DOLPHINDB_TEST_PASSWORD
Each password is resolved per connection through the credentials seam: set the
referenced name in the process environment / managed credential store, never
the value in YAML. dolphindb_switch changes active (persisted); the server
argument on dolphindb_query / dolphindb_execute overrides per call. A
profile patch may instead replace the executor row's whole config for a
fixed deployment.
dsh --profile <name> --dump-config
The output shows a # == @tradercjz/dsh-dolphindb layer mounting three rows — dolphindb (executor), tool-dolphindb (tools), dolphin-skills (skills). Starting the profile registers dolphindb_query / dolphindb_execute / dolphindb_switch and exposes the 23 dolphindb-* skills through the skill tool.
The repo type-checks and builds against a sibling deepseek-harness checkout (see tsconfig.json):
pnpm install
pnpm run typecheck # against ../deepseek-harness
pnpm run build # tsc -b + tsdown → lib/
dsh plugin add github:… installs source, so pnpm runs prepare (a self-contained tsdown transpile with no harness checkout and no type check). The first install fails until the profile's pnpm-workspace.yaml authorizes the build scripts (the dsh error prints the exact commit-pinned key — prefer that over the loose form below):
allowBuilds:
"@tradercjz/dsh-dolphindb@https://codeload.github.com/tradercjz/dsh-dolphindb/tar.gz/<commit-sha>": true
# Native file-watcher pulled transitively by the `dolphindb` client
# (dolphindb → xshell → sass → @parcel/watcher); set `false` to skip the
# native build and keep the JS fallback.
"@parcel/watcher": true
The assets/skills/ trees are vendored from the official dolphindb/DolphinX_Skill repository. One source SKILL.md (dolphindb-daily-factor) had an unquoted description that breaks YAML parsing; the vendored copy quotes it.
MIT